C H A P T E R
7
Earthquake size & descriptive earthquake
statistics
Chapter goals
• Define seismic moment, a physical measure of
earthquake size.
• Describe the origin of earthquake magnitude
scales.
• Describe commonly used magnitude scales.
• Describe the relationship between magnitude
and energy.
• Use magnitude to explore patterns in earth-
quake size and frequency of occurrence.
• Describe earthquake spatial and temporal af-
tershock patterns including magnitude.
• Discuss common earthquake catalogs.
As seismometer design and manufacture im-
proved and standardized instruments became
available, the goal of quantitatively comparing
earthquakes came within reach. Much of the
earthquake rupture process remained unclear
(the elastodynamics of faulting were not yet
fully developed) but seismogram-based mea-
surements could be compared, combined, and
cataloged. An obvious quantity to investigate
was earthquake size and if possible, estimates of
earthquake energy. More quantitative descrip-
tions of earthquake geography and frequency
were also important. Large events were obvi-
ously less frequent than small events, and after-
Foundations of Modern Global Seismology
https://doi.org/10.1016/B978-0-12-815679-7.00014-8
shock sequences were long noted and investi-
gated in terms of felt-event numbers. But with
an ability to measure seismic-wave amplitudes,
earthquake size could be quantified and com-
bined with estimated locations (see Chapter 6) to
better explore spatial and temporal earthquake
patterns.
As mentioned in Chapter 1, the modern, pre-
ferred tectonic measure of an earthquake size is
the seismic moment, M0 ,
M0 = μ A D̄ , (7.1)
where μ, the elastic shear modulus, is a measure
of the stiffness of rock surrounding the ruptured
portion of the fault, A is the area of the portion
of the fault that ruptured, and D̄ is the average
slip over the rupture surface. Seismic moment
has the intuitive interpretation that earthquake
size is a product of a measure of the rupture-
adjacent-rock’s ability to store energy, the area
of the fault that moved, and the distance the
rocks moved (one side relative to the other). The
units of seismic moment are N-m or dyne-cm,
both of which represent a force times a distance
(the lever arm). Energy has the same funda-
mental units (ML2 T −2 ) but we distinguish these
physically different quantities by using N-m for
the moment and joules for the energy. From a
dimensional-analysis viewpoint, one would ex-
pect a unit-free scale factor (possibly a combina-
197 Copyright © 2021 Elsevier Inc. All rights reserved.
198 7. Earthquake size & descriptive earthquake statistics

tion of physical parameters) to relate energy and particles, moves, and movement embodies ki-
moment. netic energy. The sum of the potential and ki-
Seismic moment is a static measure of an netic energies integrated over time equals the
earthquake’s size, a fundamental earthquake energy expended moving the ground.
property central to any description of a seismic As an example, consider a location in the
event, and a quantity relatively well-constrained ground situated directly above a monochro-
using modern seismic observations. Although matic seismic energy source. The displacement
the ground motion amplitude is directly pro- of the ground, x(t), at that position can be repre-
portional to M0 , moment is not the only source sented using
characteristic that influences seismic shaking
 
intensity. Ground-motion amplitude and dura- 2πt
tion also depend on dynamic properties that x(t) = A cos , (7.2)
T
are not included in the seismic moment. The
short-period spectral content of shaking, includ- where A is the wave amplitude and the period
ing periods that damage human-made struc- is T . The ground velocity is
tures, trigger liquefaction and landslides, etc.,
   
are strongly influenced by the spatial distribu- 2πA 2πt
tion and temporal rate of slip accumulation dur- ẋ(t) = − sin . (7.3)
T T
ing an earthquake, as well as the speed and com-
plexity of the earthquake’s rupture propagation. The kinetic energy of a unit mass at the location
Source rupture and slip characteristics are rep- of interest is given by 12 ρ ẋ 2 (t). If we average this
resented in the seismic source time-function, quantity over a complete deformation cycle, we
s(t), or its Fourier Transform, the source spec- obtain the kinetic energy density, e(ρ, A, T ),
trum, S(ω). The concepts of seismic moment and
source time functions arose from simple earth- T
quake rupture models developed after magni- 1ρ
e(ρ, A, T ) = ẋ 2 (t) dt
tudes had begun to be measured, compiled, and 2T
0
analyzed. The key point is that magnitude is a
mixture of both the static and dynamic earth-  ρ   2πA 2 T 
2πt

quake properties, which contributes to its value, = sin2 dt
2T T T
but also makes detailed interpretations of mag- 0
nitude subtle and complex.
A2
= ρ π2 . (7.4)
T2
7.1 The energy in seismic waves The energy density is proportional to the wave
amplitude squared and the frequency squared
For hazards and engineering, an ideal mea- (f = 1/T ). Thus for the same vibration ampli-
sure of earthquake size would be the energy tude, a higher-frequency signal carries more en-
transported by seismic waves, ES , as well as ergy than a lower-frequency signal. Over the
the frequency and duration of shaking. We can course of a cycle of deformation, the mean po-
quantify the energy carried by a seismic wave tential and kinetic energies are equal, so es (ρ, A,
by considering the history of a particle as it re- T ) = 2 e(ρ, A, T ).
sponds to a transient deformation. As a wave Now imagine that we have a spherical wave
passes, the particle, which has an elastic poten- source radiating energy in all directions about
tial energy by virtue of its connection to nearby the source. The wave amplitude decreases as
 7.2 Earthquake magnitude scales
the wave propagates as a result of the expan-
sion of the wavefront, and anelastic attenuation.
If we integrate our expression for the seismic-
wave energy density at one location over an en-
tire spherical wavefront and include the wave
attenuation effects, we obtain an expression of
the form
 2
A
Es = F (r, ρ, c) × , (7.5)
T The concept of earthquake magnitude, a rela-
where ES is the energy radiated by the source
as seismic waves, r is the distance the wave has
traveled, ρ is the density near the wavefront,
and c is the wave speed. The logarithm of the
energy is proportional to the logarithm of am-
plitude over period, A over T :
 
A
log Es = 2 log + log F (r, ρ, c) . (7.6)
T rupture details. Earthquakes with similar seis-
We represent the common, base-ten, logarithm
as log, and will use ln for the natural logarithm.
For a spherical wave propagating in a uniform
medium, a functional form of F (r, ρ, c) can be
derived, but the form is more complicated for
the Earth. We could relate seismic energy to seis-
mic wave amplitude and period directly using
the above equation if F (r, ρ, c) were known.
As detailed below, Richter and others defined
magnitude scales to include log (A/T ) at least
partly because the seismic-wave energy was
proportional to log (A/T ), and because ampli-
tude and period could be measured from seis-
mograms. This was a good start and the idea
worked better than expected. But construction
of a detailed relationship between magnitude
and energy remained a challenge because the
energy in the waves also depended on unknown
source physics and amplitudes depended on
propagation effects. Relating magnitude to en-
ergy was approached empirically. Gutenberg
and Richter (1956) estimated a form for F (r, ρ, c)
and showed that observed values and the early
empirical energy-magnitude relationships were
199
reasonable. But as with all things magnitude, re-
lationships are reasonably clear for first-order
comparisons, but the use of the quantities for
precise work requires a deep understanding of
the problem.
7.2 Earthquake magnitude scales
tive-size scale based on measurements of seis-
mic wave amplitudes, was developed by Kiyoo
Wadati and Charles Richter in the 1930s, more
than 30 years before the first seismic moment
was estimated in 1964. Even with paper records,
ground-motion amplitude measurements could
be made efficiently. However, traditional mag-
nitudes are limited because the character of
seismic waves is strongly influenced by fault
mic moments may have very different source-
time functions, which leads to different short-
period seismic-wave amplitudes. Even in the
simplest cases, the duration of rupture can pro-
duce frequency-dependent effects that influence
wave amplitudes in systematic ways that may
be difficult to recognize without lengthy anal-
ysis. Despite these limitations, earthquake-size
estimates based on wave amplitude measure-
ments remain useful because of their simplicity
and long history. Short-period magnitudes are
also important because high-frequency shaking
in a narrow frequency band is often responsible
for earthquake-related damage.
In the 1930’s, Richter was investigating earth-
quake processes in southern California and col-
lecting observations from a seismic network out-
fitted with Wood-Anderson seismic sensors that
had a standardized instrument response. A uni-
form seismometer sensitivity is important if you
want to compare quantitatively the ground mo-
tions from earthquakes at multiple seismic sta-
tions. When comparing measurements, Richter
noticed that for individual earthquakes, the log-
arithm of the maximum amplitude observed
200 7. Earthquake size & descriptive earthquake statistics
FIGURE 7.1 Origin of the local magnitude scale, based
on the systematic decrease of seismic amplitudes with epi-
central distance. Different symbols identify amplitudes mea-
sured from different events. The data are for Southern Cali-
fornia earthquakes in January, 1932. The dashed line shows
the formula developed by Richter to match the general trend
of the observations (from Richter, 1958).
on a Wood-Anderson seismogram seemed to
follow a systematic pattern of decreasing am-
plitude with increasing epicentral (source-to-
station) distance (Fig. 7.1). The amplitude pat-
terns suggested that a single number could, to
first order, characterize earthquake size, and led
Richter to develop the first magnitude scale.
Magnitude scales are based on two simple as-
sumptions: (1) given the same source-receiver
geometry and two earthquakes of different size,
the “larger” event will typically produce larger-
amplitude arrivals; (2) the effects of source
depth and geometric spreading and attenuation
can be removed from the amplitude measure-
ments. The general form of all magnitude-scale
formulas is
M = log (A/T ) + f (, h) + Cs + Cr , (7.7)
where A is the absolute maximum ground dis-
placement of the seismic wave on which the
scale is based, T is the period of the signal, f is
a correction for epicentral distance () and focal
depth (h), Cs is a correction for the siting of a sta-
tion (e.g., variability in amplification due to local
rock type), and Cr is a regional correction. A log-
arithmic scale is used because the seismic-wave
amplitudes of earthquakes vary enormously.
A unit increase in magnitude corresponds to a
10-fold increase in ground displacement. Magni-
tudes are averages of estimates computed from
multiple stations to overcome amplitude biases
caused by source propagation effects not ac-
counted for in the simple formula: radiation pat-
tern, directivity, and anomalous path properties.
Four primary magnitude scales are in use today:
ML , mb , MS , and MW .
7.2.1 Local magnitude (ML )
Richter’s first seismic magnitude scale was
motivated by his goal to issue the first cat-
alogue of California earthquakes. This cata-
logue contained several hundred events, whose
size ranged from barely perceptible to large,
and Richter felt that an earthquake description
should include a more objective size measure-
ment to communicate the event’s significance.
As described above, he had observed that for
measurements from uniform instruments, the
logarithm of maximum ground motion for sev-
eral earthquakes decayed with distance along
parallel curves (Fig. 7.1). He surmised that rela-
tive size of events could be calculated by com-
paring event amplitudes with amplitudes corre-
sponding to a reference event:
log A − log A0 = ML , (7.8)
where A and A0 are the displacements of the
earthquake and the reference event respectively,
both observed at a prescribed distance. Richter
chose his reference earthquake, with ML = 0,
such that A was 10−6 m at an epicentral distance
of 100 km. Using Richter’s reference event def-
inition and amplitude correction curve (log A0 ),
 7.2 Earthquake magnitude scales
we can approximate Eq. (7.8) using
⎧ different from those assumed. The discrepan-
⎪
⎪ log(A) − 0.27 + 1.65 log  ,
⎪
⎪
⎨ 0 ≤  ≤ 200 km
ML = (7.9)
⎪
⎪
⎪
⎪ log(A) − 3.55 + 3.07 log  ,
⎩
200 <  ≤ 600 km
where A is the amplitude in millimeters. At first
glance Eq. (7.9) is not in the form of (7.7), but
Richter made a number of restrictions that can
be factored out of (7.7). First, all the instruments
used were narrowband and identical, and thus
the maximum-amplitude was always of a single
dominant period, T ∼ 0.8 s. Second, all the seis-
micity was shallow (less than 15 km deep), and
the travel paths were confined to southern Cali-
fornia. Thus the corrections for regional depen-
dence and focal depth are approximately con-
stant, and Eq. (7.9) is a simplified form of (7.7).
The scale is not without issues however, for ex-
ample, the original assumed instrument gain is
now believed to have been overestimated.
Earthquakes with ML ≤ 2.5 are called mi-
croearthquakes and are rarely felt. The smallest
events that are recorded have magnitudes less
than zero, and the largest ML recorded is about
7, which corresponds to seven orders of mag-
nitude in ground displacement. In practice, ML
is usually a measure of the regional-distance S
wave. The magnitude for each of the horizon-
tal components is averaged in a least-squares
sense to give an ML for a given station. ML may
vary considerably from station to station, due
not only to station corrections but also to vari-
ability in the S-wave radiation pattern. The val-
ues of ML from all stations are averaged to give
the “magnitude.”
ML in its original form is rarely used today
because Wood-Anderson torsion instruments
are uncommon and, of course, because most
earthquakes do not occur in southern California.
Modern data can be analyzed by digitally sim-
ulating the response of the Wood-Anderson in-
strument, but doing so has led to the recognition
201
that both the period and the gains were slightly
cies are not important for a first-order use of ML
values to categorize earthquakes, but subtle seis-
mological analyses should account for details.
The modern version of ML for a region with seis-
mic wave attenuation characteristics similar to
California, recommended by the International
Association of Seismology and Physics of the
Earth’s Interior (IASPEI), one of the eight As-
sociations that comprise the International Union
of Geodesy and Geophysics (IUGG), is
ML = log A + 1.11 log R + 0.00189 R + 0.91 ,
(7.10)
where R is the hypocentral distance (R =
√
h + 2 ) where h is the hypocentral depth
2
(both in km). A is the amplitude in microns ob-
served on a horizontal-component seismogram
filtered to replicate a Wood-Anderson standard
seismograph with unit magnification. The stan-
dard Wood-Anderson short-period instrument
had a magnification thought to equal 2800, but
which was later estimated to equal 2080, see Bor-
rmann and Dewey (2014). For regions with dif-
ferent attenuation characteristics, a similar form
is recommended, but calibrated to the region.
ML remains a very important magnitude scale
because it was the first widely used earthquake
“size measure,” and all other magnitude scales
are to some extent calibrated to ML . In addition,
ML remains a very useful scale for engineering.
Many structures have natural periods close to
that of a Wood-Anderson instrument (0.8 s), and
the extent of earthquake damage is closely re-
lated to ML .
7.2.2 Body-wave magnitude
Local magnitude works fine at local and near-
regional distances, but the limitations imposed
by instrument type and the distance correction
make it impractical for global characterization
of earthquake size. Gutenberg measured magni-
tudes (mB ) of distant or teleseismic, earthquakes
202 7. Earthquake size & descriptive earthquake statistics
using the intermediate-period (roughly 5 s to
20 s) period body-wave train that included P,
PP, and S. The use of the largest observed phase
helped reduce the effects of radiation pattern
and the use of PP extended the distance range
to distances for which direct P does not exist
(beyond the P and S core shadow zones). A mod-
ern intermediate-period body-wave magnitude,
mB (BB), has been defined and utilized more re-
cently. The formula is
Vmax
mB (BB) = log + Q (h, ) ,
2π
where Vmax is the largest amplitude in an abso-
lute sense in a ground velocity seismogram (mi-
crons/s) observed in the P -wave train (largest
amplitude before P P arrives). The distance and
depth correction factor, Q (h, ) is described be-
low and is the same correction used for the
short-period body-wave magnitude, mb . Al-
though the period is not explicit in the mB (BB)
expression because it is based directly on ground
velocity, measurement of the period is encour-
aged because it contributes useful information
on the spectral content of the maximum ground
velocity.
In the early 1960’s, with the establishment
of the Worldwide-Standard Seismographic Net-
work (WWSSN), which included standardized
short- and long-period seismometers, it became
convenient to define body-wave magnitude
based on the amplitude of short-period tele-
seismic P waves. The short-period observations
were less noisy than longer-period observations
and the change led to a significant increase in
the number of small events that could be quan-
tified using teleseismic observations. The move
to shorter periods also enabled incorporation of
measurements from short-period seismic arrays
that were established around this time. These
advantages were particularly important for un-
derground nuclear explosion monitoring, which
was a primary reason the WWSSN network was
established. The resulting body-wave magni-
tude is represented using the symbol mb . Tradi-
tionally, this magnitude is based on amplitudes
measured in the first few cycles of the P -wave
arrival and is computed using
mb = log (A/T ) + Q (h, ) , (7.12)
where A is the ground-motion amplitude in mi-
crometers and T is the corresponding period in
seconds. The reason for using several cycles of
the P wave is that the effects of radiation pattern
and wave interactions with Earth’s surface for
shallow events can result in a complicated wave-
(7.11)
form. In practice, the period at which mb is usu-
ally measured is 1 s (historically, the WWSSN
and many regional network instruments had a
response “peaked” near 1 Hz). For a given event,
scatter of the order of ±0.3 mb units is not un-
usual, so robust estimates require extensive av-
eraging.
The correction for distance and depth, Q(h,
), is empirical. Fig. 7.2 includes plots of Q(h, )
correction surfaces from GR, Gutenberg and
Richter (1956), and VC, Veith and Clawson
(1972). Corrections are often shown as contour
plots, but here we show them using distance
curves for a set of sampled source depths. Each
row of the plot represents the amount added to
log (A/T ) to produce a magnitude for the cor-
responding source depth. The height of each
bar represents the size of the correction; a bar
with zero height would correspond to a mb 5.0
earthquake. The GR corrections are used by the
USGS and other agencies, the VC corrections are
used by the Comprehensive Test Ban Treaty Or-
ganization’s (CTBTO) International Data Cen-
ter (IDC). Corrections are fairly uniform in the
30◦ to 90◦ degree range, but are complicated
at closer, upper-mantle distances, which reflects
the complexity of the body waves in this epicen-
tral distance range. The correction dramatically
decreases at 20◦ because P-wave interactions with
the mantle transition zone structure can result
in very large amplitude arrivals. The correc-
tions dramatically increase near 100◦ because
diffraction along the core dramatically reduces
 7.2 Earthquake magnitude scales 203

FIGURE 7.2 The correction Q(h, ), Gutenberg and Richter (1956), and P (h, ), Veith and Clawson (1972), that are applied
to determine the body-wave magnitude (computed to adjust observations for an mb 5 reference event). The third panel shows
the difference between the others and all plots are on the same scale. The GR corrections remain in use by the USGS and other
agencies, the VC corrections are used by the CTBTO IDC. Note that in the teleseismic range (30 to 90 degrees), the curves
vary relatively smoothly. The steep increase beyond 100 degrees is an attempt to correct for the loss of short-period energy
as the wave diffracts along the core-mantle boundary. The complexities in regional distances are related to upper mantle
complexity with both depth and geographic location within the shallow Earth.

short-period P amplitudes. The VC corrections
are smoother than the GR values. For shallow
events, the corrections are similar, but include
differences up to 0.25 magnitude units. For
deeper sources, especially at regional distances,
the differences are as great as one magnitude
unit. The use of different correction surfaces
can lead to systematic differences in magnitude,
but correction surface issues are not the only
complication of working with mb . Differences
in measurement practice, including the instru-
ment/filter response used to isolate the short-
period signals on a digital seismogram, and the
length of time over which the maximum ampli-
tude is measured also affect estimated mb . The
IDC’s primary interest is in nuclear explosions,
so they restrict the mb measurement window to
the first five seconds of the P-waveform. This
is perfectly acceptable for explosions, which re-
lease their energy very quickly. However, large
earthquake rupture durations are much longer
than five seconds, and the largest amplitude
may arrive much later. The USGS and others
who monitor earthquakes use a time window
that extends as late as the arrival of PP or 60 sec-
onds to capture large amplitude P waves. For
large earthquakes, the time-window difference
is significant and the IDC mb will be a lower
bound on the USGS measurements. Perhaps the
most important take-away is that although mag-
nitude is viewed as a simple metric, the details
are rather complex and care is needed to work
with magnitudes in a detailed, quantitative way.
For most applications, where magnitude repre-
sents a rough indicator of earthquake size, these
issues are less important.
7.2.3 Surface-wave magnitude (MS )
Beyond about 600 km, the long-period seis-
mograms of shallow earthquakes are dominated
by surface waves, usually with a period of ap-
proximately 20 s. Earth’s crustal structure is
such that Rayleigh waves with periods near 20 s
204 7. Earthquake size & descriptive earthquake statistics
(roughly 18 to 22 s, sometimes called the crustal
Airy phase) propagate efficiently, which makes
them a good choice for measuring earthquake
size. However, surface-wave amplitudes de-
crease with the source depth and rather than try
to correct the surface-wave amplitude for depth,
we use surface waves only to compare shallow
earthquake size using surface-wave magnitude.
In modern notation, the classic MS is repre-
sented as MS (20), that is, the magnitude deter-
mined from the amplitude measured at 20 s.
Since surface-wave amplitudes depend on dis-
tance differently than those of body waves, we
have a different form for the distance correc-
tions, and in fact, a number of different formulas
have been proposed. The equation for comput-
ing surface-wave magnitude is
MS (20) = log A/T + 1.66 log  + 3.3 , (7.13)
where A is the amplitude in micrometers (or mi-
crons) of Rayleigh wave observations with a pe-
riod close to 20 s. In general, the amplitude of
the Rayleigh wave on the vertical component
is used for the distance range 20◦ ≤  ≤ 160◦ ,
but this has not always been the case. MS has
long been used as a better metric of the size of
larger earthquakes than mb , and the energy re-
lationship between MS and seismic energy was
used in the definition of moment magnitude,
MW . So MS (20) remains an important parameter
for many historic and modern earthquakes. The
existence of modern digital data offers oppor-
tunities to produce more robust and consistent
metrics. The broad-band MS recommended by
IASPEI has the form
Vmax
MS (BB) = log + 1.66 log  + 3.3 , (7.14)
2π
where Vmax is the maximum absolute ground
velocity in microns/s within the vertical compo-
nent Rayleigh-wave train within a period range
of 3 s ≤ T ≤ 60 s and applicable across the dis-
tance range 2◦ ≤  ≤ 160◦ .
These are not the only formulas used to es-
timate surface-wave magnitude. A similar, but
revised, distance correction is used by groups
that monitor underground nuclear explosions.
As was the case with mb , changes to magnitude
scales such as MS can provide more consistent
metrics for earthquake size suitable to applica-
tion with modern digital observations, but they
also introduce a time-variable complexity that
requires earthquake scientists to be aware of the
subtle but systematic differences between mod-
ern and historic magnitude estimates.
7.2.4 Other magnitude scales
The total number of magnitudes is substantial
and we cannot review them all. Small earth-
quakes recorded by local networks at short dis-
tances have been categorized using the duration
of the coda following the S-wave. The prac-
tice allows estimates of nearby earthquakes,
even when the amplitude is driven off-scale or
clipped by the instrument. The more qualitative
seismic intensity measurements have been cal-
ibrated with seismogram-based magnitudes so
that the maximum intensity can provide an esti-
mate of the size of large historical earthquakes.
The magnitude of large historic tsunamigenic
earthquakes have been measured using tsunami
effects calibrated by modern measurements of
magnitude and modern tsunami.
Regional magnitude, mb (Lg)
In large more stable regions of the continents
such as eastern North America, which for many
years hosted fewer seismic stations so that the
local data were less common, ML has been re-
placed with mb (Lg), a measure based on the
regional-distance Lg arrival. Lg is a superpo-
sition of shear-wave and higher mode energy
traveling with a group speed roughly in the
range from 3.2 to 3.6 km/s (regionally variable).
Lg is a mixture of energy leaving the source from
a range of directions that average some of the di-
rectional variations in wave amplitude. The U.S.
Geological Survey uses a piecewise continuous
 7.2 Earthquake magnitude scales 205

Box 7.1 Magnitudes and Explosion Monitoring

Seismic magnitudes are an integral part of modern approaches to monitoring large under-
ground nuclear explosions. We can split seismic explosion monitoring analyses into two tasks,
explosion identification and explosion characterization - magnitudes have been central to each.
mb , was in part developed for the analysis of explosion sources and smaller-magnitude earth-
quakes that were difficult to quantify using mB or MS (20) during the era of paper records
(and to some extent, even with modern digital data). After some investigation, it was discov-
ered that the ratio of short-period P -wave to intermediate-period Rayleigh-wave energy in
explosive sources differed from that typical of earthquakes. Thus examination of the ratio of
mb /MS (20) provided a way to identify candidate explosive sources from the far more numer-
ous earthquakes. For smaller explosions, often observed only at regional distances, the use of
teleseismic-based magnitudes is problematic. Regional discriminants based on the same princi-
ples but quantified using short-period spectral ratios of P -, S-, or Lg -wave amplitudes are used
in place of mb and MS .

FIGURE B7.1.1 Explosion mb -yield observations and relationships for Nevada Test Site (NTS) (left) and Semi-
palatinsk Test Site (STS) (right). Scatter in the NTS observations is partially from differences in explosion emplacement
conditions (from Maceira et al., 2017).
In addition to their use in event discrimination, approximate relationships between magnitude
and explosion yield have also been developed. P -wave-based relations are more reliable for
this purpose. In Fig. B7.1.1 we show known yield versus mb for the Nevada and Semipalatinsk
(former Soviet Union) test sites. The STS relationship is relatively consistent

mb = 4.45 + 0.75 log10 W , (B7.1.1)

where W is the explosion yield in kilotons of TNT equivalent. The yield relationship for the
Nevada Test Site is more complicated. For hard-rock sources
206 7. Earthquake size & descriptive earthquake statistics

mb = 3.92 + 0.81 log10 W . (B7.1.2)

Clearly, explosions with the same mb from each test site correspond to significantly different
yields. The systematic difference in mb for a given yield is caused by the strong attenuation
beneath the U.S. test site within the Basin and Range (underlain by a hot and attenuative
upper mantle) relative to the less attenuating upper mantle under the Soviet test site. In the
1970’s and 80’s, estimating yields from the measured magnitudes before the mantle variations
were understood caused political issues between the two superpowers. In addition to geologic
effects, inherent in these yield relationships are assumptions about the explosion depth and em-
placement conditions, which can complicate the broad use of simple relations. Transporting a
relationship from one test site to another (or even a different region of a single test site) requires
careful consideration of the geologic characteristics of both sites.

formula in the region the western conterminous U.S. is much stronger

⎧ than in the east, which leads to a more substan-
⎪
⎪ 3.75 + 0.90 log  + log(A/T ) ,
⎨
0.5◦ ≤  ≤ 4◦
mb (Lg ) =
⎪
⎪ 3.30 + 1.66 log  + log(A/T ) ,
⎩
4◦ <  ≤ 30◦
(7.15)
where A is the peak ground motion amplitude in
microns calculated using a seismogram filtered
with a short-period seismometer response filter,
T is the period in seconds, and  is the epicen-
tral distance in degrees.
The recommended IASPEI formula for
mb (Lg) is
mb (Lg ) = log A + 0.833 log R
+ 0.04343 γ (R − 10) + 2.13 , (7.16)
where A is the maximum absolute amplitude
observed on a regional seismogram filtered with
a short-period instrument response in microns,
γ is a regional attenuation parameter, and R is
the hypocentral distance. The Lg signals prop-
agate well through regions with relatively uni-
form crustal thicknesses but do not traverse re-
gions with significant changes in crustal thick-
ness, including continent-ocean transitions. The
value of γ has to be calibrated for each re-
gion of application. For example, attenuation in
tial amplitude correction in the west relative to
the east.
Seismic coda magnitude
Richter and others chose to use the ampli-
tude of the major seismic arrivals to estimate
the size of an earthquake. As described earlier
in this chapter, one of the factors that reduces
the size of the direct waves during propagation
is the scattering of the wave’s energy by het-
erogeneity in the Earth. Although P -waves scat-
tered from S-waves can arrive before the S wave,
most of the scattered energy arrives after the
major arrival because the scattered waves travel
a longer distance (from the source to the scat-
terer then to the receiver). Scattered waves fol-
low all the major seismic arrivals (P , S, Lg , etc.)
and form what we call a seismic wave’s coda.
The scattered waves that arrive closest in time
to the major arrival have the largest amplitudes
because they experience fewer scattering inter-
actions and propagated only a slightly larger
distance than the major arrival. Amplitudes de-
crease steadily as the time after the major ar-
rival increases because the later-arrival scattered
waves may have been scattered multiple times
and/or simply followed a longer path to the sta-
 7.2 Earthquake magnitude scales
tion. The result is a roughly exponential decay
in the amplitude of the signal behind the ma-
jor arrivals. We can predict the pattern theoreti-
cally (and gain insight into the nature of seismic
wave speed heterogeneity) or we can measure
the coda shape empirically, since it appears to be
a relatively stable function of frequency within a
particular region.
The coda amplitudes carry information on
earthquake size since the overall amount of scat-
tering is proportional to the size of the waves
that are being scattered. If we know the char-
acteristic shape of the coda in a particular re-
gion, then we can use the coda level to esti-
mate source size. The averaging properties of
coda smooth out focusing or defocusing effects
that can substantially increase or decrease the
peak amplitude of a seismic wave, particularly
at short periods. Also, since the coda includes
waves that leave the source in all different di-
rections they can help reduce the effect of earth-
quake radiation patterns (provided enough of
the coda to insure sufficient averaging is observ-
able). Major seismic wave amplitudes are mod-
ulated by the faulting geometry of the earth-
quake, which complicates earthquake size esti-
mates that do not include a correction for the
effect (radiations patterns are discussed in more
detail in Chapter 17). As a result of the intrin-
sic averaging of the seismic wavefield by the
scattering process, magnitudes based on cali-
brated coda-shapes can be used to estimate ro-
bust earthquake magnitudes using fewer seis-
mic stations. In practice, a coda analysis is per-
formed for a range of frequencies using narrow-
band coda signals to construct an estimate of
the source spectrum (Fig. 7.3). The magnitude
is then estimated from the source spectrum. The
source spectrum is discussed in more detail in
the next Chapter. Coda magnitudes are an im-
portant tool for constraining earthquake size
and spectra, but the effects at short periods and
short coda durations needed to model close,
small-magnitude earthquake may require cali-
207
FIGURE 7.3 Example shear-wave narrowband-filtered
coda-envelopes (solid lines) estimated by smoothing the log-
arithmic average of the codas of the two horizontal compo-
nents. Dashed curves show synthetic envelope fits for the
coda amplitude measurements. After a short time, the loga-
rithm of the coda amplitude decreases linearly with distance.
Note that the higher-frequency coda attenuates faster be-
cause it corresponds to signals with shorter wavelengths that
make it more susceptible to intrinsic attenuation and scat-
tering. The differences in the levels of the coda correspond
to source properties or near-station geology (from Mayeda,
2003).
bration to account for lateral variations in scat-
tering.
7.2.5 Magnitude saturation
Both MS (20) and mb were designed to be as
compatible as possible with ML . Sometimes all
three magnitudes have the same value - but, this
is not always the case. For small earthquakes
differences may be caused by signal-to-noise
issues. For example, MS (20) is hard to mea-
sure for small or deep earthquakes. For large
earthquakes, when signal-to-noise is not an is-
sue, magnitude differences reflect earthquake
physics not well understood at the time that the
magnitude scales were developed. Each of ML ,
mb , and MS (20) depends on an amplitude mea-
surement at a different period (and wave type),
roughly 0.8 s (usually S), 1.0 s (usually P), and
20 s (Rayleigh), respectively. If earthquakes ra-
208 7. Earthquake size & descriptive earthquake statistics

FIGURE 7.4 Illustration of magnitude saturation for the high-frequency magnitude mb versus the lower-frequency magni-
tudes MW and the start of saturation for MS at large magnitudes. The data are from the U.S. Geological Survey measurements
contained in the ISC catalog. For several of the earlier large earthquakes, the GCMT MW is used since the values in the ISC
catalog reflect initial underestimated values. The data are from the years 2004-2017 and the symbol shading indicates the
number of observations at each coordinate pair. The darker symbols indicate more data, lighter symbols indicate fewer ob-
servations.

diated seismic waves with a uniform amplitude
at all periods, the magnitudes would agree as
long as distance and depth corrections were ac-
curate. However, earthquakes radiate uniform
amplitudes as a function of period only for peri-
ods shorter than roughly π times the earthquake
duration, τr . A rough estimate of a typical earth-
quake duration in seconds can be computed us-
ing
τr [s] ∼ 100.6 M−2.8 .
This value is approximate, because the spread of
durations for any particular magnitude is easily
a factor of 2 of this “typical” value, so do not
use the relation for precise work. For compar-
ing magnitudes, we must be concerned about a
decrease in seismic wave amplitude that occurs
when the measurement period is shorter than
roughly ∼ 3 τr .
Physically, what happens is that longer pe-
riod components of the signal, which corre-
spond to waves with wavelengths much larger
than the spatial dimensions of the rupture and
periods that smooth out temporal variations in
the rupture process, do not sense any rupture
complexity. Small earthquakes have short dura-
tions and ML , mb , and MS (20) are all measured
at periods longer than 3 τr and the magnitudes
generally agree. But for large earthquakes, the
duration of the event increases first beyond the
short-periods used to measure ML and mb and
the magnitudes stop increasing with earthquake
size because the short-period peak amplitudes
stop increasing as rupture grows and complex-
ity causes complex interference patterns among
(7.17)
the shorter period components of the seismo-
gram. We call this effect magnitude saturation.
Classic ML and mb saturate roughly near
magnitude 6.5 (which has a duration of roughly
10 s). MS (20) remains a reasonably consistent
measure of earthquake size until the duration
of the event is roughly a minute, and it begins
to saturate for magnitudes just larger than 8.0.
The effect is illustrated in Fig. 7.4, which is a
comparison of USGS mb estimates with MW es-
timates on the left and USGS MS (20) with MW
estimates on the right. Ideally, the magnitudes
would agree and follow the lines, which have
a slope of unity. The saturation effect is quite
clear for mb and although the number of large
 7.3 Seismic energy, magnitude, and moment magnitude
events is limited, saturation is also apparent
for MS (20). Note also that the relationship for
smaller magnitudes is not simple either, neither
mb nor MS follow the one-to-one line with MW .
Broad-band magnitude, mB (BB) and MS(BB)
are measured using a broader frequency ranges
than the classic magnitudes. As a result, they
saturate at higher magnitudes than their clas-
sic counterparts, but they too saturate. The first
step towards a non-saturating magnitude scale
was an empirical relation between magnitude
and energy.
7.3 Seismic energy, magnitude, and
moment magnitude
Gutenberg and Richter constructed empirical
relationships between mB and MS and ES in ergs
(1 erg = 10−7 joules):
log Es = 5.8 + 2.4 mB , (7.18)
and
log Es = 11.8 + 1.5 MS . (7.19)
Obviously, the energy calculation in Eqs.
(7.18) and (7.19) suffers from all the problems
of the magnitude estimation. In particular, E
estimated using a saturated magnitude will be
underestimated. Eq. (7.19) is fairly robust, since
MS does not saturate until large magnitudes.
Eq. (7.19) provides an interesting insight into the
tremendous range of earthquake size. The differ-
ence between the energy released in an MS = 7.0
and an MS = 8.0 earthquake is a factor of 101.5 ,
or ∼ 32. In other words, the seismic energy re-
leased in a magnitude 8.0 earthquake is over 30
times greater than that released in a magnitude
7.0 earthquake, and it is three orders of mag-
nitude greater than that released in a MS = 6.0
earthquake.
Kanamori et al. (1993) related ES computed
from local and regional distance seismograms
with ML in southern California. Their analysis
209
suggests that for small and moderate-magnitude
earthquakes in southern California, ML is a sta-
ble estimator of earthquake energy. The relation-
ship, valid for ML ≤ 6.5, is
log Es = 9.05 + 1.96 ML , (7.20)
where ES is in ergs. The relationship has a slope
of 2, which indicates energy in the short-period
(T ∼ 0.8 s) range used to estimate ML , increases
by a factor of 100 for a unit increase in ML . This
differs with the usual relationship between MS
and energy for large earthquakes because MS
and ML are measured at different frequencies
and the source spectrum of earthquakes is fre-
quency dependent. The ML expression is not
valid for large earthquakes because ML satu-
rates because of the same source-spectrum fre-
quency dependence. For first-order approxima-
tions, the MS relationship is usually the one cho-
sen, but the important point is that the complex-
ity of earthquake rupture can complicate magni-
tude’s interpretation. Using magnitude for other
than direct event comparisons requires under-
standing of the history, limitations and subtleties
of this important and oft-quoted quantity.
We can also relate seismic moment to seis-
mic energy. An earthquake’s energy budget can
be quantified in terms of the change in stress
caused by the fault rupture and offset. Kostrov
(1974) showed that the radiated seismic energy
is proportional to the rupture area, fault offset,
and the stress decrease on the fault produced by
the earthquake (usually called the stress drop):
1
Es ≈ σ D̄A , (7.21)
2
or, rearranging terms using the definition of
M0 = μAD̄,
σ
Es ≈ M0 . (7.22)
2μ
We can use this expression to relate M0 to magni-
tude through Eq. (7.19). If we assume that earth-
quake stress drop is roughly constant and equal
210 7. Earthquake size & descriptive earthquake statistics
to about 3 MPa, this yields the relation
log M0 = 1.5 MS + 16.1 , (7.23)
where the moment is in units of dyne-cm. This
equation gives a simple way to relate magnitude
to seismic moment and, in fact, can be used to
define a new magnitude scale, MW , called mo-
ment magnitude:
log M0 − 16.1
Mw = . (7.24)
1.5
The moment-magnitude scale, derived by
Kanamori (1977), is calibrated to MS but will not
saturate because M0 does not saturate. Gener-
ally, estimating M0 is more complicated than es-
timating magnitude, but modern seismic analy-
ses routinely provide M0 estimates for all events
larger than MW ∼ 5.0. Using the non-saturating
MW magnitude scale, the largest earthquake
recorded was the 1960 Chilean earthquake, with
MW = 9.5.
7.4 Descriptive earthquake statistics
Once earthquake size was quantified, obser-
vations from around the world could be com-
pared and contrasted and patterns identified.
Just as earthquake locations greatly refined ev-
idence that earthquakes were spatially concen-
trated in relatively narrow regions, important
patterns were identified in earthquake tempo-
ral, spatial, and magnitude distributions. With
the tools of instrument-based earthquake lo-
cation and magnitude estimation, the patterns
could be investigated more broadly and more
quantitatively because analyses could be ex-
tended to smaller and more remote earthquake
sequences, which greatly increased the observa-
tions of earthquake processes.
7.4.1 The Gutenberg-Richter
relationship
For example, there is an important scaling
relationship between earthquake size and fre-
quency of occurrence. Richter and Gutenberg
first proposed that in a given region and for a
given period of time, the frequency of occur-
rence can be represented by
log N = A − b MS (7.25)
where N is the number of earthquakes with
magnitudes in a fixed range around magnitude
MS , and A and b are constants. The numerical
value of A varies with the length of time over
which seismicity is examined, the area of the
region studied, and the area’s seismicity rate.
Mathematically, A represents the logarithm of
the number of M = 0 events. For practical rea-
sons the lower magnitude threshold for analysis,
Ma , is larger than zero, and we replace MS with
(MS − Ma ) in the relation. Then a is the num-
ber of events at the magnitude threshold). Often,
N (M) represents the number of events equal to
or larger than M. The form of the relationship
remains the same, but the constant, a, must be
interpreted differently. The constant b in (7.25) is
called the b value and often, if not typically, has
a value of 1.0. We can rearrange terms in (7.25)
and substitute M0 from (7.23) for MS to obtain
−(b/1.5)
N (M0 ) = A M0 . (7.26)
This type of power-law size distribution arises
from earthquake self-similarity. Self similar-
ity has a number of interpretations, but gen-
erally means that small and large earthquakes
are similar. That means that the ratio of fun-
damental physical quantities such as fault slip
to fault length or fault slip to fault area scale
in proportion to event size. The same ideas
lead to a roughly constant earthquake stress
drop (actually stress drop varies over a limited
range). Fig. 7.5 is a histogram of the number
 7.4 Descriptive earthquake statistics
FIGURE 7.5 Number of earthquakes as a function of the
logarithm of seismic moment (or magnitude). This is a global
data set for all earthquakes since 1977. The solid line is a
straight line fit to the central part of the distribution. The
equation of the line is log N = 19.9 − 0.65 log M0 , which has a
slope close to −2/3, and a b-value for the magnitude depen-
dence close to 1.0.
of earthquakes as a function of seismic mo-
ment and magnitude. The data set includes
over 50,000 events with magnitudes greater than
about MW 5 that have occurred since 1977. In
the intermediate size range (where most of the
earthquakes contribute), the b value is roughly
1.0. In general, b values are between 23 and
1 and do not show much regional variability.
If we consider Eq. (7.25) determined per year,
then the A value gives the maximum expected
earthquake (assuming b = 1). Globally, we ex-
pect roughly one earthquake per year for which
MW ≥ 8.0. A b-value of unity implies that each
year we should observe roughly 10 or more
events for which MW ≥ 7.0 (see Figure in chap-
ter 1) and 100 or more events for which MW ≥
6.0. With this in mind, you can understand the
seismologist’s frustration with the question, “Is
this the big one?” which is heard any time an
earthquake does damage in California. Earth-
quakes of magnitude 6.0 are hardly unusual
phenomena, at least globally.
The one case in which the b-value departs
significantly from ∼ 1.0 is earthquake swarms.
Swarms are sequences of earthquakes that are
211
clustered in space and time and are not associ-
ated with an identifiable mainshock. The b val-
ues for swarms can be as large as 2.5, which im-
plies that no large earthquakes accompany the
occurrence of small-magnitude events. Swarms
most commonly occur in volcanic regions, and
the generally accepted explanation is that faults
simply are not large or continuous in this en-
vironment and that stress is substantially het-
erogeneous. Thus, the maximum moment ex-
pected for a given earthquake is small, and
many smaller events must accommodate the
strain accumulation (since the energy associated
with a magnitude of 5.0 is 1000 times smaller
than that for a magnitude of 7.0, we mean many
smaller events!)
Eq. (7.25) is often used in seismic hazard anal-
ysis to determine the maximum credible earthquake
during a specified time window. If we assume
that A and b values determined for a given
time range are self-similar in both size and fre-
quency of occurrence, the Gutenberg-Richter re-
lation can be extrapolated to larger time win-
dows. The recurrence time of an earthquake of
magnitude M is proportional to N −1 . For exam-
ple, if we monitored a region for one year and
found that b = 1 and A = 3.5, we then would
expect a magnitude 4.5 earthquake in the next
10 years and a magnitude 5.5 earthquake in the
next century. Obviously, this type of analysis is
loaded with assumptions; a maximum magni-
tude will eventually be reached, and a 1-yr win-
dow may poorly represent the earthquake fre-
quency of occurrence for all magnitudes.
We can use Eq. (7.25) to estimate the yearly
energy release from earthquakes. The largest
event in a given year usually accounts for ap-
proximately 10% of the total seismic energy re-
lease, and the events with magnitudes greater
than 7.0 account for more than 10% of the total.
Proceeding this way, the annual energy release
by earthquakes averages 1.0 − 2.0 × 1017 J/yr.
The average energy release is smaller than that
released in large megathrust earthquakes. The
1960 Chilean megathrust earthquake had a seis-
212 7. Earthquake size & descriptive earthquake statistics
mic moment of roughly 2 × 1023 N-m. Using
Eq. (7.22) and assuming a 3 MPa stress drop and
shear modulus of 40 GPa, the estimated energy
release is about 80 × 1017 J, at least 40 times the
annual average. The annual earthquake energy
release is comparable to that expected from a
magnitude in the range from 8.3 to 8.5. Earth-
quake energy release is relatively small com-
pared to the energy release of other geophysical
processes. The 1991 eruption of Mt. Pinatubo in
the Philippines alone released roughly 8 × 1019 J.
The heat flux out of the solid Earth is roughly 44
terawatts, or about 1.4 × 1021 J/year. For com-
parison, global human energy consumption is
roughly 6 × 1020 J/yr.
7.4.2 Earthquake occurrence rates
Estimation of the probability of an earth-
quake occurrence is central to seismic hazard
estimation, which is a fundamental problem for
earthquake science. In many physical systems,
random events are modeled as a Poisson pro-
cess - events are assumed to occur randomly,
but at a uniform rate. Under such assumptions
probabilities of event occurrence can be esti-
mated using a Poisson distribution, which is
parameterized completely by the rate of event
occurrence. For earthquakes in general, the as-
sumption of a Poisson process is inappropri-
ate. Earthquakes occur in sequences and many
events are not independent. But a Poisson model
has value when we consider the largest earth-
quakes that may occur in a particular region.
If these events can be considered independent,
their occurrence rate can be estimated using
earthquake history, earthquake catalogs, and
paleo-seismic investigations. The results can be
used to estimate earthquake occurrence proba-
bilities. More generally, earthquake occurrence
patterns are viewed as the sum of two com-
ponents, background earthquakes that occur
as part of a tectonically driven Poisson pro-
cess, and triggered earthquakes that are initi-
ated by other earthquakes. The separation of
earthquakes into background and triggered ac-
tivity requires separating or de-clustering events
in an earthquake catalog. Many approaches for
earthquake catalog de-clustering have been de-
veloped - early approaches were heuristic, more
recent approaches have incorporated statistical
models of earthquake occurrence.
On a global scale, we can crudely assume
that most large earthquakes are mainshocks, not
triggered by other events (not completely accu-
rate, but for a simple experiment, it’s a reason-
able starting point). Fig. 7.6 shows the cumula-
tive distribution of the number of large events
per year observed for the last ∼ 120 years. We
used the USGS PAGER catalog updated with the
USGS Common Catalog for more recent years
for the computation. We chose to define large
with a minimum magnitude of 7.0, but the dif-
ference between an M 6.9 and an M 7.0 earth-
quake is of no physical significance. The number
of cataloged events from 1900 to 2018 with M ≥
7 is 1872. The cumulative large-event count-per-
year distribution is consistent with our estimates
of the mean number of earthquakes per year in
Chapter 1 - roughly 15-16 (we can consider the
mean value as an estimate of the mean of the dis-
tribution). The match to the distribution is good,
and the pattern is reasonably consistent consid-
ering that we did not remove aftershocks, which
we know do not follow the same Poisson-rate
assumptions as mainshocks.
To refine the analysis, we applied a very sim-
ple, heuristic declustering algorithm to the orig-
inal earthquake list. We assumed that any event
occurring within four weeks and 200 km of an
earlier large event is an aftershock. Applying
this rule to the original 1872 large events iden-
tified 164 (roughly 10%) of the events as sus-
pected aftershocks. The cumulative histogram of
the de-clustered events is compared with a Pois-
son distribution parameterized by 14 events per
year in Fig. 7.7. The largest visual change in the
histogram is the removal of any year with more
than 28 events (four of the original 32 events in
1943 were classified as aftershocks). The cumu-
 7.5 Patterns in earthquake sequences 213

FIGURE 7.6 Cumulative distribution of the number of years (count) versus the number of large, M ≥ 7, in each year. The
dotted curve shows the predictions for a Poisson distribution with a global large earthquake rate between 15-16 events per
year.

FIGURE 7.7 Cumulative distribution of the number of years (count) versus the number of large, M ≥ 7, in each year
with suspected aftershocks removed. The dotted curve shows the predictions for a Poisson distribution with a global large
earthquake rate between 14 events per year.

lative Poisson distribution fit is slightly better
for the de-clustered data. But we must keep in
mind that we employed a simple de-clustering
rule. Changing the rule to classify events in
larger or smaller temporal and spatial ranges
as aftershocks would affect the pattern. A de-
crease in the mean number of events per year
by roughly one event per year (from 15-16 to
14) is more consistent with de-clustered cumu-
lative distribution. Of course any estimate of av-
erage number of large earthquakes in a given
year should include the aftershocks, so the num-
ber of 15-16 remains the appropriate answer to
the general question of how many large earth-
quakes typically occur each year? The larger the
events we want to analyze, the fewer data we
have and the less certain the analysis. Using the
Onur and Muir-Wood historical catalog of giant
earthquakes (estimated MW ≥ 8.8), from about
1550 to the present, a similar analysis suggests a
mean of roughly 5-6 giant earthquakes per cen-
tury. Assuming a Poisson distribution, this sug-
gests a 55% chance of experiencing five or fewer
and an 70% chance of experiencing six of fewer
of these giant events in any hundred-year inter-
val.
The global pattern of large earthquakes ap-
pears roughly compatible with a Poisson de-
scription, but this is far from a complete char-
acterization of earthquake occurrence patterns.
A substantial number of earthquakes are trig-
gered by other earthquakes and are not indepen-
dent in their timing, location, or size. For these
events, the uniform-rate assumption associated
with a classic Poisson distribution is inapplica-
ble.
7.5 Patterns in earthquake sequences
Earthquakes occur in sequences. Nearly all
large earthquakes are followed by a sequence
of smaller earthquakes, known as aftershocks,
which are spatially and temporally related to
the large event. We assume that the same pro-
cess also occurs for even the smallest events, but
those aftershocks are too small for routine de-
tection. Some events are preceded by seismicity,
214 7. Earthquake size & descriptive earthquake statistics
which after the fact are called foreshocks. The
largest earthquake in the sequence is called the
mainshock, but these labels are only certain af-
ter the sequence is complete. The terminology is
valuable to enable discussion of events as they
occur, but must remain flexible as a result of the
diversity of earthquake sequence types. Much of
this text can be considered an analysis of main-
shock properties - in this section we discuss fore-
shock and aftershock processes and their associ-
ated patterns.
7.5.1 Foreshock patterns and earthquake
nucleation
Many earthquakes appear to be preceded by
detected precursory seismic activity called fore-
shocks and the percentage is even larger (over
70%) for large earthquakes (for which foreshock
activity may be easier to detect). Foreshocks
themselves are no different from other earth-
quakes, so identifying them as foreshocks is the
challenge. Given the complexity of earthquake
rupture, it makes some sense that fault ruptures
would start with precursory activity. The 1960
Chile earthquake began with a sequence of ma-
jor earthquakes that may have been triggered
by precursory aseismic slip. The 33 hours prior
to the great event included at least three ma-
jor events. Although at a smaller scale, several
recent large earthquakes also began with ex-
tended sequences of foreshocks, including the
2011 Tohoku and the 2014 Northern Chile earth-
quakes (Fig. 7.8). Just over three weeks before
the nucleation of the MW 9.0 Tohoku earthquake,
precursory activity began near and migrated to-
wards the large event’s hypocenter. Two days
before the great event, an MW 7.3 event rup-
ture a region close to the Tohoku earthquake’s
hypocenter. The 2014 Tarapacá, Chile (MW 8.1)
earthquake showed a similar pattern - it was
preceded by nearly two weeks of moderate to
large offshore earthquakes in a region that had
not registered a large earthquake since a very
large event in 1877.
However, earthquake sequences quite simi-
lar to the recent foreshock sequences have been
observed in other regions believed to be large-
earthquake capable, but no large earthquake fol-
lowed. Thus, it remains difficult to distinguish
elevated earthquake activity that occurs prior to
a larger event from elevated earthquake activ-
ity that is not followed by a large event. The
interpretation of seismic activity as precursory
remains ambiguous. In the case of Tohoku, the
earliest activity could easily have been inter-
preted as foreshocks of the MW 7.0 event, not
the larger MW 9.0 event. One research avenue
that might reduce the ambiguity is work to col-
lect and study information on possible aseismic
deformation that may drive foreshock activity.
Since most events are offshore, that effort will re-
quire marine deformation networks, which are
under development but require substantial re-
sources. In addition, other observations, such as
those obtained before the 2004 Parkfield earth-
quake, suggest that expecting a simple, clear
pattern may not be realistic. But even an ap-
proach that works part of the time could have
substantial value. Given the complexity of earth-
quake precesses, earthquake scientists must con-
tinue to inform nonspecialists that any notable
earthquake in a region is a indication of a poten-
tial for future shaking and a reminder for those
that live in earthquake-prone regions to stay pre-
pared.
7.5.2 Aftershock patterns and rupture
area
Awareness of aftershocks must be very old,
since large earthquakes without aftershocks are
exceptionally rare. But the term aftershock orig-
inated in descriptions of the 1755 Lisbon earth-
quake. The largest earthquake in a sequence,
known as the mainshock, introduces a nearby
strain change that perturbs adjacent segments of
the same fault as well as the complex system
of faults generally found in all fault systems.
Faults and fault segments adjacent to the rup-
 7.5 Patterns in earthquake sequences 215

FIGURE 7.8 Plate boundary thrust sequences for the 2011 Mw 9.0 earthquake in Tohoku, Japan (A), and the 1 April 2014
Mw 8.1 earthquake in north Chile (B). Shaded dots indicate great earthquake epicenters. The fault geometries of large fore-
shocks are indicated by focal mechanisms. Arrows indicate migration direction. In the maps on the right of each panel, shade
denotes depth of foreshocks (from Brodsky and Lay (2014)).

ture respond to stress changes induced by the
strain change, and aftershocks are produced.
Aftershocks typically begin immediately after a
mainshock and are distributed throughout the
volume around the rupture affected by the strain
adjustment (the source volume). Aftershocks can
be dangerous as a result of the damage to struc-
tures caused by the mainshock. But in most
earthquake sequences, the sum of the aftershock
seismic moments rarely exceeds 10% of the
mainshock moment. For a typical shallow inter-
plate MS = 7.0 earthquake, thousands of small
aftershocks may occur. In general, for shallow
earthquakes, the largest aftershock magnitude is
roughly a unit smaller than the mainshock (1.2
units smaller on average, this is called Båth’s
Law). These patterns vary with tectonic environ-
ment. Oceanic transform earthquakes, oceanic
intraplate earthquakes, and deep earthquakes
(earthquakes that rupture within oceanic litho-
sphere) produce fewer aftershocks than shallow
continental and interplate events, and the largest
aftershock in a sequence in the oceanic litho-
sphere is more often roughly two magnitude
units smaller than the mainshock.
Typically, the relative frequency of after-
shocks decays rapidly at first, then slowly tran-
sitions into the normal background rate. Omori
studied aftershocks in Japan in the late 1800’s
and early 1900’s and developed an empiri-
cal formula to describe typical aftershock rates
(Omori’s Law). Let n(t) represent the frequency
of aftershocks at time t after the mainshock, then
C
n(t) = , (7.27)
(K + t)P
where K, C, and P are constants that depend on
the size of the earthquake. The P value is usu-
ally close to 1.0-1.4. The key features of the curve
are a rapid decrease in the number of events
followed by a very slow return to background
activity levels. Fig. 7.9 is a plot of the time his-
tory of the aftershocks of the 2015 Nepal earth-
quake. The first large event in the sequence be-
216 7. Earthquake size & descriptive earthquake statistics
gins a classic Omori decay until a second cluster
of events occurs two weeks later, which starts a
second Omori decay pattern. Some recent stud-
ies have suggested that the decay rates for af-
tershocks depends on the rate of tectonic strain
loading in the source volume. In this model, re-
gions that accumulate strain more slowly host
longer duration aftershock sequences. The im-
plication is that relatively long aftershock se-
quences should be expected in regions of slow
strain accumulation, such as the interiors of con-
tinents. At least some well known earthquake
sequences are consistent with the idea.
Aftershock spatial distributions are often used
to infer the mainshock rupture area. For most
earthquakes, the rupture area (or aftershock
area) scales with magnitude. Utsu and Seki
(1954) developed an empirical relationship
log A ∼ 1.0 MS + 2.0, (7.28)
where A is measured in m2 . Aftershock zones ex-
pand slightly during a month or more following
the mainshock presumably involving changes
in fluid pressures or aseismic deformation oc-
curring in response to the mainshock. For this
reason, when possible, rupture area is estimated
from the extent of the aftershock zone after 1
to 2 days. Fig. 7.10 includes map and cross-
sectional views of the aftershocks of the 1983
Borah Peak, Idaho, earthquake. The aftershocks
define a zone approximately 70 km long and a
dipping surface compatible with the southwest-
dipping nodal plane of the focal mechanism.
Note that the mainshock hypocenter is located
at the southeast end and near the bottom of the
rupture. Apparently the Borah Peak earthquake
rupture mode was unilateral, the rupture propa-
gated up and to the northwest. It is common for
the hypocenter of a mainshock to be located near
the bottom of the rupture; likely related to fault
strength conditions in the crust.
Earthquake hypocenters outside subduction
and collision zones are seldom deeper than
15-20 km. Along any fault, the deepest earth-
quakes define the base of the seismogenic zone.
Within the seismogenic zone the crust deforms
primarily either by stable or unstable sliding on
faults and more rarely by brittle failure when
subjected to stresses greater than the strength of
the material. Rock strength and frictional behav-
ior depend on deformation rate, temperature,
pressure, and composition. For most materials
that could realistically make up a significant
fraction of the crust and for realistic tempera-
ture profiles, rock strength increases to a depth
of roughly 15 km and then decreases rapidly.
Near the same depth frictional properties transi-
tion from those conducive to unstable sliding to
those favoring stable sliding. The precise depth
varies from place-to-place with details of com-
position and thermal structure. The maximum
in the strength is often near the base of the un-
stable seismogenic zone. The geometry of the
seismogenic zone has important consequences
for earthquake size. The implication is that a
maximum fault width is available for rupture,
below which stable sliding or ductile deforma-
tion occur. Earthquakes that rupture this entire
zone can be classified as large, and those that
only rupture part of the seismogenic zone can
be classified as small.
Thus far we have described fault slip, D, in
terms of average slip on the part of the fault
that ruptured. Within an earthquake rupture,
slip varies considerably, and in fact the slip of-
ten appears to be concentrated on patches called
asperities (by analogy to the microstructural
protrusions in Amonton’s early work on fric-
tion). Asperities are generally thought to repre-
sent zones of relatively high stress drop. Average
stress drop appears independent of earthquake
size, but parts of a rupture may have stress
drops an order of magnitude higher than the
average. The variation in frictional properties
or slip history on the fault responsible for the
nonuniform slip and/or stress drop remains an
active area of research. Aftershocks on the fault
hosting the rupture often concentrate around the
edges of the mainshock asperities. Fig. 7.11 is a
map of interplate aftershock density in the re-
 7.6 Earthquake catalogs 217

FIGURE 7.9 (left) Mainshock and aftershock locations for the 25 April, 2015 Gorkha, Nepal Earthquake (MW 7.9). (right)
Magnitude timeline and Omori-Law summary using the number of earthquakes in successive 12-hour time intervals. The
aftershocks decay according to Omori’s law until a large aftershock occurred just over two weeks later.

gion surrounding the 2011 MW 9.0 earthquake
off the coast of Honshu, Japan. Regions of rel-
atively intense aftershock activity surround the
substantial area of mainshock slip centered on
the star. Many additional aftershocks occurred
within the upper and lower plates above and
below the region of significant mainshock slip,
but slip between the plates in that region was
subdued.
7.6 Earthquake catalogs
Now that we have a basic understanding of
how earthquakes are located, and how magni-
tude is used to estimate an earthquake’s size, we
can discuss the archives of this information, seis-
mic catalogs. In a sense, an earthquake catalog is
a table of information about earthquakes (when
and where they occurred, how big they were,
etc.). The most commonly used information in
a catalog are the hypocentral locations and ori-
gin times of earthquakes, but seismic catalogs
also include the arrival-time, amplitude and pe-
riod measurements used to estimate the source
parameters. The arrival time measurements are
the basis of tomographic travel-time based in-
vestigations of Earth’s interior. Modern catalogs
also include information on the earthquake size
(seismic moment) and faulting geometry of the
source if it is available, so are a rich source of in-
formation for tectonic investigations. The avail-
ability of faulting geometry information also al-
lows seismologists to use the waveforms to con-
strain Earth’s Interior by modeling seismograms
directly.
The first lists of earthquakes (we know of)
were constructed in the 1600’s. Karl Ernst Adolf
von Hoff of Germany (1771-1837) began pub-
lishing annual lists of worldwide earthquakes
in 1826. The first global earthquake catalogs re-
lied on historic accounts of shaking and earth-
quake damage. French seismologist, Alexis Per-
rey (1807-1882), compiled local, regional, and
global earthquake lists. In the 1850’s, Robert
Mallet, assisted by his college-student son, John
William (later a chemistry professor at the Uni-
versity of Alabama), compiled a comprehensive,
global earthquake catalog for the British As-
sociation. Mallet (1810-1881), an Irish engineer
and one of the founders of earthquake seismol-
ogy, published broadly on earthquakes and used
field observations of the 1857 Neapolitan Earth-
quake to investigate the origin and mechanics
of earthquakes, seismic waves, and the damage
they cause.
218 7. Earthquake size & descriptive earthquake statistics

FIGURE 7.10 Aftershocks of the 1983 Borah Peak, Idaho, earthquake. The fault ruptured the surface for about 40 km. Note
that the aftershocks define a plane that is consistent with the N22◦ W plane of the focal mechanism (from Stein and Barrientos,
1985).

None of these earlier researchers had the
benefit of data from seismometers, only his-
toric records. Seismometer development began
in the early 1800’s and usable instruments that
became available in the late 1800’s allowed de-
tection of more small earthquakes. In the late
1800’s and early 1900’s Fernand Jean Baptiste
Marie Bernard comte de Montessus de Ballore
(1851-1923) constructed perhaps the largest his-
torical earthquake catalog and included some
information from seismometer deployments.
Volumes of his 21,000 event catalog once occu-
pied 85 feet of shelf space in the Societe de Ge-
ographie in Paris (Davison, Charles. 1927. The
Founders of Seismology. Cambridge [England]:
The University Press).
Seismicity, in the modern sense, is a term
used to indicate the spatial, temporal, and size
distributions of earthquake activity in a given
region. Gutenberg and Richter (1949, 1954) com-
 7.6 Earthquake catalogs
FIGURE 7.11 Map view of interplate aftershock density
recorded throughout the one-year period immediately fol-
lowing the 2011 Tohoku-Oki mainshock. Typical reference
focal mechanism are shown in the lower-right corners. The
earthquake generated may aftershocks within the upper
plate above the mainshock rupture, the focus here is on in-
terplate events along the plate boundary. The black star indi-
cates the epicenter of the 11 March 2011 MW 9.0 Tohoku-Oki
mainshock. Diamonds represent the epicenters of small re-
peating earthquakes. The gray curve identifies the down-dip
limit of interplate seismicity. Large aftershock locations are
identified. Most of the interplate aftershocks lie outside the
region of large slip during the mainshock (modified from
Kato and Igarashi, 2012).
pleted a comprehensive summary of earthquake
geography including seismogram-based earth-
quake magnitude estimates. Their efforts estab-
lished a number of regional seismicity classifi-
cations, and earthquake-depth categories (shal-
low, intermediate, and deep) that we continue
to use. Their classic summary, Seismicity of the
Earth, (1949, 1954) is available online in PDF for-
mat, and along with Richter’s Book Elementary
219
TABLE 7.1 U.S. Geological
Survey earthquake descriptors.
Magnitude Range Descriptor
M ≥ 8.0 Great
7.0 ≤ M < 8.0 Major
6.0 ≤ M < 7.0 Strong
5.0 ≤ M < 6.0 Moderate
4.0 ≤ M < 5.0 Light
3.0 ≤ M < 4.0 Minor
2.0 ≤ M < 3.0 Very Minor
Seismology (1958), tabulated and summarized
much ground-breaking research of the 1930’s
40’s and 50’s. Between 2007 and 2013, the U.S.
Geological Survey produced a set of posters
with a Seismicity-of-the-Earth theme. You can
find and download them online.
As discussed in Chapter 1, earthquakes are
commonly classified based on size of the event.
Table 7.1 shows the U.S. Geological Survey
earthquake descriptors based on magnitude.
The descriptors can be used to quickly con-
vey the potential impact an earthquake without
specificy the exact magnitude.
7.6.1 Modern earthquake catalogs
Modern earthquake catalogs are for the most
part available online. No catalog contains all the
information on all the earthquakes that occur.
We say a catalog is complete to a specified mag-
nitude threshold if it has all the earthquakes of
that size or larger. Even above the completeness
threshold, the amount of information included
in a catalog varies. Some catalogs are more suit-
able for some purposes than others. Catalogs
serve users with a broad range of interest, from
the public, to the press, and researchers. Many
of the online interfaces provide a range of in-
formation formats that include different levels
of information. Over long periods of time, cat-
alogs are merged, revised, and sometimes re-
designed. Many of the hypocenter and magni-
tude estimates for historical earthquakes come
220 7. Earthquake size & descriptive earthquake statistics
from analysis by Gutenberg and Richter in the
1950’s (Abe, 1975; Pacheco and Sykes, 1992).
The USGS Common Catalog. The USGS
maintains a number of catalogs including a real-
time update of their Common Catalog, COM-
CAT. The term common indicates that the cat-
alog integrates results obtained from the USGS
global and regional networks. For example, an
earthquake in east-central California may be lo-
cated by three regional networks, northern and
southern California, and Nevada as well as the
global monitoring network. Some coordination
is required to insure that duplicate events are
not included and that the best estimate of the
hypocenter parameters are archived. The infor-
mation in COMCAT is dynamic and events are
added and revised in near real time. During the
weeks that follow an event, revisions may be
performed as the data are reviewed (large events
are more completely analyzed quicker).
COMCAT includes reasonably complete in-
formation on global activity with magnitudes
greater than about 4.5. For smaller events, the
results are probably complete for earthquakes
in the conterminous US for earthquakes with
magnitudes greater than 3.0. Regions with dense
seismic networks such as California, Puerto
Rico, etc include events complete to smaller
magnitude ranges. The U.S. is not unique, and
many countries and regions produce catalogs
of smaller earthquakes that occur within their
regions of focus. A thorough study including
information on small earthquakes in a specific
region always benefits from some communi-
cation with any local groups monitoring local
activity. The UK’s International Seismic Center
(ISC) has historically aggregated many of the
catalog results into the ISC Catalog. The task is
substantial, particularly assessing the quality of
solutions and arrival-time picks, but the result is
a valuable resource.
The Global Centroid Moment Tensor Cata-
log. As mentioned in Chapter 1, researchers ini-
tially at Harvard and then Columbia University
have developed the Global Centroid Moment
Tensor Catalog (GCMT) - a valuable resource
that includes not only consistent estimates of
more than 50,000 post-1976 earthquake seismic
moments, but also important information on the
faulty geometry of the events. The GCMT is a
major resource for seismologists and other geo-
scientists studying earthquakes and tectonics.
The information is publicly available through
a search interface or the entire catalog can be
downloaded for more custom analyses. For a
complete appreciation of the catalog informa-
tion, details on the methods used to estimate the
event parameters are documented in a series of
scholarly articles (Dziewonski et al., 1981; Ek-
ström et al., 2012).
The International Seismic Center, Global-
Earthquake-Model Catalog. ISC-GEM is a his-
toric earthquake catalog constructed to help the
GEM group estimate seismic hazard and risk
(Storchak et al., 2013). ISC-GEM includes many
older events that have been relocated using
modern algorithms and earth models to create
a more consistent set of earthquake hypocenter
and magnitude estimates. The variability in the
quality and quantity of seismic travel time and
amplitude measurements results in a variable,
albeit improving quality of solutions with time.
Since the seismological community continues to
analyze information on historic and older earth-
quakes, the GEM catalog will continue to evolve
for some time. Although global appears in the
name, information from regional networks is
integrated into the GEM catalog, so in some lo-
cales, the magnitude threshold is lower than in
other global historic catalog.
The USGS Centennial Catalog. The USGS
Centennial catalog is similar to the ISC-GEM, in
that older events, with magnitudes larger than
roughly 5.0 are reanalyzed to be more consistent
with their modern equivalents. Again, the qual-
ity and quantity of the older data remains lower
than their modern equivalents, so not all historic
events can be located with the accuracy and pre-
cision of more recent earthquakes.
 7.7 Summary
Catalogs of Large Historic Events. Onur
and Muir-Wood (2014) constructed a catalog
of the largest earthquakes (M ≥ 8.8) believed
to be complete back to 1700 and nearly com-
plete back to the late 1500’s. The catalog in-
cludes 18 events since 1700. From the informa-
tion available, South America is the region most
prone to the largest earthquakes. Paleoseismic
methods have been used to construct paleo-
catalogs of earthquakes in many regions. The re-
sults are much less certain than instrumentally-
based results, but provide an important look
into the deeper history of fault behavior. No sim-
ple global catalog of all these events has been
compiled - but histories for various subduc-
tion zone segments (Cascadia, Chile, etc.) have
been estimated using geologic evidence related
to tsunami and submarine landslide deposits.
Trenching across and along continental faults
also provides information on large fault systems
such as the San Andreas in California, North
Anatolian in Turkey, and many others.
Catalogs of Earthquake Impact. Several cata-
logs include information on the impact of earth-
quakes (and tsunami) on humans. Fatality cata-
logs spanning more than one century have been
developed by Utsu (2002). The USGS’ PAGER
catalog was developed to use for rapid hazard
estimation following large threatening modern
events (Allen et al., 2009). The catalog was last
updated in 2008. NOAA’s National Centers for
Environmental Information maintains catalogs
of significant historic and modern earthquakes
(deadly and damaging) and tsunami (NOAA,
2018). As with the hypocenter catalogs, precise
casualty numbers are not available for all deadly
earthquakes and the range of estimates for in-
dividual catastrophes can be dramatic. For ex-
ample, the 2010 Haiti earthquake death toll esti-
mates range from 90,000 to 300,000.
All seismic catalog information has uncer-
tainty and as new information is uncovered and
analyzed, some seismic catalogs are often up-
dated. As more data are converted from pa-
per records to digital, location analyses can be
221
performed using consistent and modern earth
models and magnitude formulas. This trend will
likely continue for some time. Perhaps most im-
portant is the need to understand the history
of an earthquake catalog and recognize changes
in procedures used to locate and estimate the
size of earthquakes, as well as acknowledge the
history of seismic networks, which generally
evolved to include more observations during the
last century. Catalogs are an important source of
information on background earthquake occur-
rence and earthquake sequences.
7.7 Summary
Richter developed magnitude with a mod-
est goal of classifying earthquake into three
strength levels (large, medium, and small). As
measurements accumulated, he and Gutenberg
realized that it worked much better than ex-
pected. In the years that followed, they extended
the idea repeatedly and used magnitude to iden-
tify important patterns in earthquake size distri-
bution and mapped earthquake geography with
a more quantitative basis than ever before. They
also empirically related magnitude to earth-
quake energy release, a fundamental advance
in earthquake physics. But magnitude measures
a complicated mix of static and dynamic ef-
fects associated with earthquake rupture, and
the long history of measurement has resulted
in a metric of earthquake size that is both ex-
tremely valuable for basic size classification, but
that also includes subtle dependencies on the
diversity of earthquake rupture characteristics.
Using magnitude quantitatively requires care,
but the success of the scale for communicating
a first-order estimate of an earthquake size is
without peer in seismology.
Earthquake sequences exhibit patterns, but
not all follow the same script. Omori provided a
quantitative model of the rate of the aftershock
occurrence that indicates an initial rapid decay
followed by an extended return to background
222 7. Earthquake size & descriptive earthquake statistics
levels. We have seen enough foreshock activity
to know that while they are not as ubiquitous
as aftershocks, they are often observed. Some
optimism is warranted that a combination of
geodetic and seismic observations may provide
information on impending earthquakes in some
(but not all) instances. But knowing when activ-
ity is foreshock in nature remains a challenge.
The observations suggest that short-term earth-
quake prediction remains an elusive goal. With
more than a century of instrumental seismicity
recorded in catalogs, including more than 50,000
consistently measured size and faulting geom-
etry estimates, seismologists and other earth-
quake scientists have a wealth of information
to continue studying the complex and diverse
earthquake processes.